any floating point library with more exponent bits?

[hrbigelow]

hi,
i was wondering if anyone has ever seen a floating point library that
uses at least 20 bits to represent the exponent, rather than the 11
bits for a double-precision standard IEEE-754 float. thanks in
advance!

henry

 

[Gordon Burditt]

i was wondering if anyone has ever seen a floating point library that

uses at least 20 bits to represent the exponent, rather than the 11
bits for a double-precision standard IEEE-754 float. thanks in
advance!

IEEE extended-precision seems to use 15 bits for the exponent
(including exponent sign). That gets up to 1.1e+4932 or so.
I don't know of any hardware that uses more (not to say there
isn't any). Software emulation is possible (but likely slow).

Gordon L. Burditt

 

[hrbigelow]

thanks! do you know of any software emulation package that is good,
but short of "arbitrary precision" library? i guess what i'm after is
"extended range", rather than "extended precision". and, if there is a
free library, i'd rather use that than re-tool my algorithm (a bayesian
network, which alternately sums, multiplies and divides numbers, and
would be very hard to re-write in log-space).

thanks again,

henry

 

[azi.stdout]

Gnu MP Bignum Library perhaps : http://www.swox.com/gmp/

 

[jacob navia]

(e-mail address removed) a écrit :

hi,
i was wondering if anyone has ever seen a floating point library that
uses at least 20 bits to represent the exponent, rather than the 11
bits for a double-precision standard IEEE-754 float. thanks in
advance!

henry

lcc-win32 offers a 350 bits library

htpp://www.cs.virginia.edu/~lcc-win32

 

[hrbigelow]

thanks gordon. i'm actually writing in o'caml so unfortunately the
long double is not available. so far i'm using gmp as a software
emulation. any idea how much faster hardware is from software?

thanks again,

henry

 

[hrbigelow]

thanks for the excellent tip. i'm now using gmp and so far works great.